Crashes are the Steam game bug players mention most in launch reviews. In 27,156 English reviews from the first 14 days of 95 September 2026 launches, 350 mentioned a crash, in 63 of the games. The bugs that make a review most negative are rarer: a game that will not start, a lost save, a softlock. And most bug talk stays vague.
Which bugs do Steam players name in launch reviews?
The sample is every September 2026 launch in the October edition of our Steam Launch Bug Index with at least 200 reviews, 95 games. On 6 November 2026 we pulled up to 300 English reviews per game from its first 14 days on sale, 27,156 reviews in all. Word lists sorted them into 21 bug types, and a single review could count toward several.
| Bug type | Reviews | Games (of 95) | Negative | Written before 2 h of play |
|---|---|---|---|---|
| Crash, crash to desktop | 350 | 63 | 52.9% | 22.0% |
| Low FPS, poor optimization | 248 | 59 | 45.6% | 22.2% |
| Stutter, hitching | 133 | 38 | 41.4% | 24.8% |
| Softlock, quest or progression blocker | 128 | 49 | 61.7% | 14.1% |
| Freeze, hang | 84 | 40 | 47.6% | 15.5% |
| UI, menus, text, typos | 83 | 42 | 47.0% | 14.5% |
| NPC and enemy behaviour | 61 | 29 | 50.8% | 9.8% |
| Clipping, collision, physics | 57 | 26 | 49.1% | 17.5% |
| Multiplayer, disconnects, servers | 55 | 28 | 49.1% | 18.2% |
| Controller, gamepad | 54 | 37 | 40.7% | 20.4% |
| Visual glitches, flicker | 44 | 28 | 52.3% | 20.5% |
| Lost or broken saves | 43 | 33 | 62.8% | 18.6% |
| Won’t launch, black screen, stuck loading | 40 | 25 | 67.5% | 22.5% |
| Keybinds, rebinding, mouse | 37 | 23 | 48.6% | 37.8% |
| Steam Deck, Linux, Proton | 32 | 22 | 21.9% | 12.5% |
| Audio bugs | 31 | 18 | 64.5% | 38.7% |
| Memory leak, VRAM, GPU errors | 18 | 11 | 44.4% | 22.2% |
| Resolution, ultrawide, fullscreen | 18 | 15 | 44.4% | 38.9% |
| Localization, translation | 6 | 5 | 50.0% | 16.7% |
| Achievements not unlocking | 4 | 4 | 50.0% | 25.0% |
| Anti-cheat, DRM, third-party launcher | 2 | 2 | 2 of 2 | 0 of 2 |
| Any of the 21 types | 1,271 | 92 | 49.0% | 21.2% |
| All reviews in the sample | 27,156 | 95 | 15.8% | 16.1% |
Crashes lead by a wide margin. They show up in two thirds of the launches, and 37 of the 95 games had three or more crash reviews in two weeks. Performance complaints come next, then stutter: hitches during play, counted apart from a low frame rate overall.
The bottom of the table matters as much as the top. Players almost never write about achievements, anti-cheat or translation in their first two weeks. Low counts there say more about what players reach in two weeks than about whether those systems work.
Most reviews that talk about bugs never get as specific as the table. 2,055 reviews used the words bug, buggy, glitch or broken, and 1,580 of them (76.9%) named none of the 21 types. “Buggy” in a review tells a team almost nothing it can fix. The repro has to come from testing before launch, and our Next Fest demo QA checklist is laid out by the same areas as this table, from first launch and saves to controls and localization.
Repro steps can also come from someone who has never seen the build. In a Smoke Test, one of our testers spends 8 hours on it the way a new player would: a clean install, input, the first hour of play, saving and a quit-and-relaunch, then the opening again on a lower-end PC. Crashes, blocks and freezes come back with repro steps, in 2 business days.
Which bug types turn a review negative?
The ones that stop play or erase it. Across the sample, 15.8% of reviews were negative. Reviews about a game that would not start were negative 67.5% of the time, 4.3 times the average. Audio bugs came in at 64.5%, lost saves at 62.8% and softlocks at 61.7%, each about four times the average. Audio is the odd one out. It rests on only 31 reviews, and in our hand check 15 of 20 were real audio bugs.
At the other end, Steam Deck and Linux reviews that mention a problem were negative only 21.9% of the time. Controller complaints ran at 40.7% and stutter at 41.4%. More than half of the players who wrote about those bugs still recommended the game.
The split is older than this sample. A Queen’s University team coded 472 Steam reviews by hand for a study in Empirical Software Engineering (Lin, Bezemer, Zou and Hassan, 2018 preprint, data from 2016). Bugs in negative reviews were mostly incompatibility, crashes and bugs that blocked play. Bugs in positive reviews were mostly performance, audio or visual issues and crashes. They also found 42% of bug-mentioning reviews were positive. In our 2026 count, 51.0% of reviews naming a bug type still recommended the game.
That gives the first rule for test order. Start with the bugs that end a session or wipe it, because those are the reviews least likely to forgive. The full order is in the table further down.
Which bugs do players hit before 2 hours?
Settings-screen bugs. Across all reviews, 16.1% were written before 2 hours of play. For keybind complaints the share was 37.8%, for resolution and display 38.9%, and for audio 38.7%. The median keybind reviewer had played 3.5 hours, against 7.1 hours for the whole sample. Softlocks run the other way, at 14.1%, and NPC behaviour at 9.8%: players have to get deeper into the game to hit them. Two hours is also the line for a no-questions Steam refund, and in our cut of launch reviews 30% of the negative ones were written before it.
What should you test first, given these counts?
The order below weighs three things from the data: how negative a bug type runs, how many of the 95 games had it, and how early players meet it. The weighting is an editorial call. Each row points to the block of our checklists that covers it and says whether a Smoke Test reaches it.
| Order | Bug types | What to check before release | Where it is in our lists | In a Smoke Test |
|---|---|---|---|---|
| 1 | Won’t launch, crash at start | Install from Steam on a PC that never ran the game, first boot at default resolution, missing redistributables | First launch block; redistributables guide | Yes |
| 2 | Lost or broken saves | Force-close during a save, relaunch and Continue, load a save from the previous build | Saves block; save compatibility | Yes, cloud sync on two PCs excluded |
| 3 | Softlocks | Skip, die and do tutorial steps out of order; walk the critical path; try to get stuck in geometry | First 30 minutes block | First hour only |
| 4 | Crash mid-game, freeze | A 60-minute session in one sitting, crash reporting switched on and read | Performance block | Partly, within 8 hours |
| 5 | Keybinds, resolution, audio | Rebinding survives a restart; 1080p, 1440p, 4K and ultrawide; a slider for music, effects and voice | Controls block | Yes |
| 6 | Low FPS, stutter, memory | Heaviest scene on minimum spec, first run on a cold shader cache, memory over a long session | Hardware block | On a second, lower-end PC |
| 7 | Controller, Steam Deck | Xbox and PlayStation pads through Steam Input, hot-plug, Deck boot and text size | Controls block | Controllers yes, Deck no |
| 8 | Multiplayer | Join, rejoin after a disconnect, host leaving mid-session | Multiplayer block | No |
| 9 | NPCs, physics, visuals, text, localization, achievements | Areas you changed late, text in the longest language, achievements unlocking | Changed areas block; localization block | Placeholder text yes, the rest no |
Crashes sit in rows 1 and 4 on purpose. A crash at the first boot reaches the player before anything else, and in our table it falls under the most negative type. A crash after an hour is the most common complaint, so it needs a long session rather than a quick boot test.
Rows 8 and 9 sit low only because few reviews named them. A co-op game should move multiplayer to the top, since one broken lobby or one desync between players can stop a whole group from playing. If your reviews are already in and you are sorting them, our piece on how a Steam game recovers from Mixed reviews covers what moved the score for September launches.
How did we count, and where can the count be wrong?
Each bug type is a list of English words and phrases. For types where the word alone is often praise, such as controller, Steam Deck, resolution or audio, a review only counts if a problem word (broken, can’t, doesn’t work, missing and similar) sits within 60 characters in the same sentence. Phrases like “no crashes” or “haven’t had a single stutter” are dropped.
We then read 20 random matches per type by hand, or all of them for the smallest types. Of 368 matches, 277 (75%) described a real problem of that type in the game. Performance, keybinds and physics scored 17 of 20. Multiplayer scored lowest, 12 of 20, because “rubber banding” also describes racing AI and “disconnection” also describes how a player feels about the story. The misses were things like time that “freezes at night” by design, a meteor “crashing down”, or players saying stutter had improved.
Four other limits apply. The reviews are English only and from players who bought the game on Steam. Steam returns the newest reviews first, and 70 of the 95 games hit our 300-review cap, so for those games the sample leans toward the second week. A review counts once per type, however many times it repeats the word. And the count shows what players write about, which is not the same as how often each bug happens. Results are aggregates; no game is named.
You can check our lists yourself. The CSV with all 21 word lists holds the patterns, the problem-word filter, the per-type counts and the hand-check scores, under CC BY 4.0 like the rest of Steam in numbers. We plan to rerun it on a later month of launches and compare. If you spot a type we sort badly, write to contact@jumbojelly.com.